How we can protect truth in the age of misinformation | Sinan Aral

Translator: Ivana Korom Reviewer: Krystian Aparta

翻译人员: Wanting Zhong 校对人员: Yolanda Zhang

So, on April 23 of 2013,

2013 年 4 月 23 日，

the Associated Press put out the following tweet on Twitter.

美联社在推特上发布了 这样一条推文：

It said, "Breaking news:

“突发新闻：

Two explosions at the White House

白宫发生两起爆炸，

and Barack Obama has been injured."

巴拉克·奥巴马受伤。”

This tweet was retweeted 4,000 times in less than five minutes,

在不到五分钟的时间里， 这条推文被转发了四千次，

and it went viral thereafter.

随后也在网络上被疯传。

Now, this tweet wasn't real news put out by the Associated Press.

不过，这条推文并不是 美联社发布的真实新闻。

In fact it was false news, or fake news,

事实上，这是一则不实新闻， 或者说是虚假新闻，

that was propagated by Syrian hackers

是由入侵了美联社推特账号

that had infiltrated the Associated Press Twitter handle.

的叙利亚黑客扩散的。

Their purpose was to disrupt society, but they disrupted much more.

他们的目的是扰乱社会， 但他们扰乱的远不止于此。

Because automated trading algorithms

因为自动交易算法

immediately seized on the sentiment on this tweet,

立刻捕捉了这条推文的情感， 【注：机器学习中对主观性文本的情感分析】

and began trading based on the potential

并且根据美国总统在这次爆炸中

that the president of the United States had been injured or killed

受伤或丧生的可能性，

in this explosion.

开始了交易。

And as they started tweeting,

而当他们开始发推时，

they immediately sent the stock market crashing,

股市迅速随之崩盘，

wiping out 140 billion dollars in equity value in a single day.

一日之内便蒸发了 1400 亿美元的市值。

Robert Mueller, special counsel prosecutor in the United States,

美国特别检察官罗伯特·穆勒

issued indictments against three Russian companies

起诉了三家俄罗斯公司

and 13 Russian individuals

以及十三个俄罗斯人，

on a conspiracy to defraud the United States

指控他们干预 2016 年美国总统大选，

by meddling in the 2016 presidential election.

合谋诓骗美国。

And what this indictment tells as a story

而这次起诉讲述的

is the story of the Internet Research Agency,

是互联网研究机构的故事，

the shadowy arm of the Kremlin on social media.

即俄罗斯政府在社交媒体上 布下的影影绰绰的手腕。

During the presidential election alone,

仅在总统大选期间，

the Internet Agency's efforts

互联网机构就

reached 126 million people on Facebook in the United States,

影响了 1.26 亿名 美国 Facebook 用户，

issued three million individual tweets

发布了 300 万条推文，

and 43 hours' worth of YouTube content.

以及 43 个小时的 Youtube 内容。

All of which was fake --

这一切都是虚假的——

misinformation designed to sow discord in the US presidential election.

通过精心设计的虚假信息， 在美国总统大选中播下不和的种子。

A recent study by Oxford University

牛津大学最近的一项研究显示，

showed that in the recent Swedish elections,

在近期的瑞典大选中，

one third of all of the information spreading on social media

在社交媒体上传播 的关于大选的信息中，

about the election

有三分之一

was fake or misinformation.

是虚假或谬误信息。

In addition, these types of social-media misinformation campaigns

另外，这些通过社交媒体 进行的误导活动

can spread what has been called "genocidal propaganda,"

可以传播所谓的“种族清洗宣传”，

for instance against the Rohingya in Burma,

例如在缅甸煽动对罗兴亚人的迫害，

triggering mob killings in India.

或者在印度引发暴徒杀人。

We studied fake news

我们在虚假新闻变成热点之前

and began studying it before it was a popular term.

就开始了对虚假新闻的研究。

And we recently published the largest-ever longitudinal study

最近，我们发表了一项 迄今最大型的关于虚假新闻

of the spread of fake news online

在网络传播的纵向研究，

on the cover of "Science" in March of this year.

在今年三月登上了《科学》期刊封面。

We studied all of the verified true and false news stories

我们研究了推特上传播的所有

that ever spread on Twitter,

核实过的真假新闻，

from its inception in 2006 to 2017.

范围是自 2006 年推特创立到 2017 年。

And when we studied this information,

在我们研究这些讯息时，

we studied verified news stories

我们通过六家独立的 事实核查机构验证，

that were verified by six independent fact-checking organizations.

以确认新闻故事的真实性。

So we knew which stories were true

所以我们清楚哪些新闻是真的，

and which stories were false.

哪些是假的。

We can measure their diffusion,

我们可以测量 这些新闻的扩散程度，

the speed of their diffusion,

扩散速度，

the depth and breadth of their diffusion,

以及深度与广度，

how many people become entangled in this information cascade and so on.

有多少人被卷入这个信息级联。 【注：人们加入信息更具说服力的团体】

And what we did in this paper

我们在这篇论文中

was we compared the spread of true news to the spread of false news.

比较了真实新闻和 虚假新闻的传播程度。

And here's what we found.

这是我们的研究发现。

We found that false news diffused further, faster, deeper

我们发现，在我们研究 的所有新闻类别中，

虚假新闻都比真实新闻传播得

and more broadly than the truth

in every category of information that we studied,

更远、更快、更深、更广，

sometimes by an order of magnitude.

有时甚至超出一个数量级。

And in fact, false political news was the most viral.

事实上，虚假的政治新闻 传播速度最快。

It diffused further, faster, deeper and more broadly

它比任何其他种类的虚假新闻

than any other type of false news.

都扩散得更远、更快、更深、更广。

When we saw this,

我们看到这个结果时，

we were at once worried but also curious.

我们立刻感到担忧， 但同时也很好奇。

Why?

为什么？

Why does false news travel so much further, faster, deeper

为什么虚假新闻比真相

传播得更远、更快、更深、更广？

and more broadly than the truth?

The first hypothesis that we came up with was,

我们想到的第一个假设是，

"Well, maybe people who spread false news have more followers or follow more people,

“可能传播虚假新闻的人 有更多的关注者，或者关注了更多人，

or tweet more often,

或者发推更频繁，

or maybe they're more often 'verified' users of Twitter, with more credibility,

或者他们中有更多 推特的‘认证’用户，可信度更高，

or maybe they've been on Twitter longer."

或者他们在推特上的时间更长。”

So we checked each one of these in turn.

因此，我们挨个检验了这些假设。

And what we found was exactly the opposite.

我们发现，结果恰恰相反。

False-news spreaders had fewer followers,

假新闻散布者有更少关注者，

followed fewer people, were less active,

关注的人更少，活跃度更低，

less often "verified"

更少被“认证”，

and had been on Twitter for a shorter period of time.

使用推特的时间更短。

And yet,

然而，

在控制了这些和很多其他变量之后，

false news was 70 percent more likely to be retweeted than the truth,

虚假新闻比真实新闻 被转发的可能性高出了 70%。

controlling for all of these and many other factors.

So we had to come up with other explanations.

我们不得不提出别的解释。

And we devised what we called a "novelty hypothesis."

于是，我们设想了一个 “新颖性假设”。

So if you read the literature,

如果各位对文献有所了解，

it is well known that human attention is drawn to novelty,

会知道一个广为人知的现象是， 人类的注意力会被新颖性所吸引，

things that are new in the environment.

也就是环境中的新事物。

And if you read the sociology literature,

如果各位了解社会学文献的话，

you know that we like to share novel information.

你们应该知道，我们喜欢分享 新鲜的信息。

It makes us seem like we have access to inside information,

这使我们看上去像是 能够获得内部消息，

and we gain in status by spreading this kind of information.

通过传播这类信息， 我们的地位可以获得提升。

So what we did was we measured the novelty of an incoming true or false tweet,

因此我们把刚收到的真假推文

compared to the corpus of what that individual had seen

和用户前 60 天内 在推特上看过的语库比较，

in the 60 days prior on Twitter.

以衡量刚收到的推文的新颖度。

But that wasn't enough, because we thought to ourselves,

但这还不够， 因为我们想到，

"Well, maybe false news is more novel in an information-theoretic sense,

“可能在信息论的层面 虚假新闻更加新颖，

but maybe people don't perceive it as more novel."

但也许在人们的感知里， 它并没有很新鲜。”

So to understand people's perceptions of false news,

因此，为了理解 人们对虚假新闻的感知，

we looked at the information and the sentiment

我们研究了对真假推文的回复中

contained in the replies to true and false tweets.

包含的信息和情感。

And what we found

我们发现，

was that across a bunch of different measures of sentiment --

在多种不同的情感量表上——

surprise, disgust, fear, sadness,

惊讶，厌恶，恐惧，悲伤，

anticipation, joy and trust --

期待，喜悦，信任——

false news exhibited significantly more surprise and disgust

对虚假新闻的回复里 明显表现出了

in the replies to false tweets.

更多的惊讶和厌恶。

And true news exhibited significantly more anticipation,

而对真实新闻的回复里，

joy and trust

表现出的则是

in reply to true tweets.

更多的期待、喜悦，和信任。

The surprise corroborates our novelty hypothesis.

这个意外事件证实了 我们的新颖性假设。

This is new and surprising, and so we're more likely to share it.

这很新鲜、很令人惊讶， 所以我们更可能把它分享出去。

At the same time, there was congressional testimony

同时，在美国国会两院前 进行的国会作证

提到了机器人账号（注：一种使用 自动化脚本执行大量简单任务的软件）

in front of both houses of Congress in the United States,

looking at the role of bots in the spread of misinformation.

在传播虚假信息时的作用。

So we looked at this too --

因此我们也对这一点进行了研究——

we used multiple sophisticated bot-detection algorithms

我们使用多个复杂的 机器人账号探测算法，

to find the bots in our data and to pull them out.

寻找并提取出了 我们数据中的机器人账号。

So we pulled them out, we put them back in

我们把机器人账号移除， 再把它们放回去，

and we compared what happens to our measurement.

并比较其对我们的测量 产生的影响。

And what we found was that, yes indeed,

我们发现，确实，

bots were accelerating the spread of false news online,

机器人账号加速了 虚假新闻在网络上的传播，

but they were accelerating the spread of true news

但它们也在以大约相同的速度

at approximately the same rate.

加速真实新闻的传播。

Which means bots are not responsible

这意味着，机器人账号

for the differential diffusion of truth and falsity online.

并不是造成网上虚实信息 传播差距的原因。

We can't abdicate that responsibility,

我们不能推脱这个责任，

because we, humans, are responsible for that spread.

因为要对这种传播负责的， 是我们人类自己。

Now, everything that I have told you so far,

对于我们大家来说 都很不幸的是，

unfortunately for all of us,

刚刚我告诉各位的一切

is the good news.

都是好消息。

The reason is because it's about to get a whole lot worse.

原因在于，形势马上要大幅恶化了。

And two specific technologies are going to make it worse.

而两种特定的技术 会将形势变得更加糟糕。

We are going to see the rise of a tremendous wave of synthetic media.

我们将会目睹 一大波合成媒体的剧增。

Fake video, fake audio that is very convincing to the human eye.

虚假视频、虚假音频， 对于人类来说都能以假乱真。

And this will powered by two technologies.

这是由两项技术支持的。

The first of these is known as "generative adversarial networks."

其一是所谓的“生成对抗网络”。

This is a machine-learning model with two networks:

这是一个由两个网络组成 的机器学习模型：

a discriminator,

一个是判别网络，

whose job it is to determine whether something is true or false,

负责分辨样本的真假；

and a generator,

另一个是生成网络，

whose job it is to generate synthetic media.

负责产生合成媒体。

So the synthetic generator generates synthetic video or audio,

生成网络产生 合成视频或音频，

and the discriminator tries to tell, "Is this real or is this fake?"

而判别网络则试图分辨， “这是真的还是假的？”

And in fact, it is the job of the generator

事实上，生成网络的任务是

to maximize the likelihood that it will fool the discriminator

尽可能地欺骗判别网络， 让判别网络误以为

它合成的视频和音频

into thinking the synthetic video and audio that it is creating

is actually true.

其实是真的。

Imagine a machine in a hyperloop,

想象一台处于超级循环中的机器，

trying to get better and better at fooling us.

试图变得越来越擅长欺骗我们。

This, combined with the second technology,

第二项技术， 简而言之，

which is essentially the democratization of artificial intelligence to the people,

就是在民众中 的人工智能的民主化，

the ability for anyone,

即让任何人

without any background in artificial intelligence

不需要任何人工智能或

or machine learning,

机器学习的背景，

to deploy these kinds of algorithms to generate synthetic media

也能调用这些算法 生成人工合成媒体。

makes it ultimately so much easier to create videos.

这两种技术相结合， 让制作视频变得如此容易。

The White House issued a false, doctored video

白宫曾发布过一个 虚假的、篡改过的视频，

of a journalist interacting with an intern who was trying to take his microphone.

内容为一名记者和一个试图抢夺 他的麦克风的实习生的互动。

They removed frames from this video

他们从视频中移除了一些帧，

in order to make his actions seem more punchy.

让他的行动显得更有攻击性。

And when videographers and stuntmen and women

而当摄影师和替身演员

were interviewed about this type of technique,

被采访问及这种技术时，

they said, "Yes, we use this in the movies all the time

他们说，“是的，我们经常 在电影中使用这种技术,

to make our punches and kicks look more choppy and more aggressive."

让我们的出拳和踢腿动作 看上去更具打击感，更加有气势。”

They then put out this video

他们于是发布了这个视频，

and partly used it as justification

将其作为部分证据，

to revoke Jim Acosta, the reporter's, press pass

试图撤销视频中的记者， 吉姆·阿考斯塔

from the White House.

的白宫新闻通行证。

And CNN had to sue to have that press pass reinstated.

于是 CNN 不得不提出诉讼， 要求恢复该新闻通行证。

There are about five different paths that I can think of that we can follow

我能想到我们可以走 的五条不同道路，

to try and address some of these very difficult problems today.

以试图解决当今我们面对 的这些异常艰难的问题。

Each one of them has promise,

每一种措施都带来希望，

but each one of them has its own challenges.

但每一种也有其自身的挑战。

The first one is labeling.

第一种措施是贴上标签。

Think about it this way:

可以这么想：

when you go to the grocery store to buy food to consume,

当你去超市购买食品时，

it's extensively labeled.

食品上会有详细的标签。

You know how many calories it has,

你可以得知它有多少卡路里，

how much fat it contains --

含有多少脂肪——

and yet when we consume information, we have no labels whatsoever.

然而当我们摄取信息时， 我们没有任何标签。

What is contained in this information?

这个信息中含有什么？

Is the source credible?

其来源是否可信？

Where is this information gathered from?

这个信息是从哪里收集的？

We have none of that information

在我们摄取信息时，

when we are consuming information.

我们并没有以上的任何信息。

That is a potential avenue, but it comes with its challenges.

这是一种可能的解决办法， 但它有自身的挑战。

For instance, who gets to decide, in society, what's true and what's false?

比如说，在社会中， 有谁能决定信息的真伪？

Is it the governments?

是政府吗？

Is it Facebook?

是 Facebook 吗？

Is it an independent consortium of fact-checkers?

是由事实核查机构 组成的独立联盟吗？

And who's checking the fact-checkers?

谁又来对事实核查机构 进行核查呢？

Another potential avenue is incentives.

另一种可能的解决手段是奖励措施。

We know that during the US presidential election

我们知道，在美国总统大选期间，

there was a wave of misinformation that came from Macedonia

有一波虚假信息来源于马其顿，

that didn't have any political motive

他们没有任何政治动机，

but instead had an economic motive.

相反，他们有经济动机。

And this economic motive existed,

这个经济动机之所以存在，

because false news travels so much farther, faster

是因为虚假新闻比真相传播得

and more deeply than the truth,

更远、更快、更深，

and you can earn advertising dollars as you garner eyeballs and attention

你可以使用这类信息 博取眼球、吸引注意，

with this type of information.

从而通过广告赚钱。

But if we can depress the spread of this information,

但如果我们能抑制 这类信息的传播，

perhaps it would reduce the economic incentive

或许就能在源头减少

to produce it at all in the first place.

生产这类信息的经济动机。

Third, we can think about regulation,

第三，我们可以考虑进行监管，

and certainly, we should think about this option.

毫无疑问，我们应当考虑这个选项。

In the United States, currently,

现在，在美国，

we are exploring what might happen if Facebook and others are regulated.

我们在探索当 Facebook 和其它平台 受到监管时，会发生什么事情。

While we should consider things like regulating political speech,

我们应当考虑的措施包括： 监管政治言论，

labeling the fact that it's political speech,

对政治言论进行标签，

making sure foreign actors can't fund political speech,

确保外国参与者无法资助政治言论，

it also has its own dangers.

但这也有自己的风险。

For instance, Malaysia just instituted a six-year prison sentence

举个例子，马来西亚刚刚颁布法案， 对任何散布不实消息的人

for anyone found spreading misinformation.

处以六年监禁。

And in authoritarian regimes,

而在独裁政权中，

these kinds of policies can be used to suppress minority opinions

这种政策可以被利用 以压制少数群体的意见，

and to continue to extend repression.

继续扩大压迫。

The fourth possible option is transparency.

第四种可能的解决方法是透明度。

We want to know how do Facebook's algorithms work.

我们想了解 Facebook 的算法是怎样运作的。

How does the data combine with the algorithms

数据是怎样与算法结合，

to produce the outcomes that we see?

得出我们看到的结果？

We want them to open the kimono

我们想让他们开诚布公，

and show us exactly the inner workings of how Facebook is working.

为我们披露 Facebook 内部 具体是如何运作的。

And if we want to know social media's effect on society,

而如果我们想知道 社交媒体对社会的影响，

we need scientists, researchers

我们需要科学家、研究人员

and others to have access to this kind of information.

和其他人能够入手这种信息。

But at the same time,

但与此同时，

we are asking Facebook to lock everything down,

我们还要求 Facebook 锁上一切，

to keep all of the data secure.

保证所有数据的安全。

So, Facebook and the other social media platforms

因此，Facebook 和其他社交媒体平台

are facing what I call a transparency paradox.

正面对我称之为的“透明性悖论”。

We are asking them, at the same time,

我们要求他们

to be open and transparent and, simultaneously secure.

在开放、透明的同时 保证安全。

This is a very difficult needle to thread,

这是非常艰难的挑战，

but they will need to thread this needle

这些公司必须直面挑战，

if we are to achieve the promise of social technologies

才能在实现社交科技承诺的同时

回避它们带来的危害。

while avoiding their peril.

The final thing that we could think about is algorithms and machine learning.

我们能想到的最后一个解决手段是 算法和机器学习。

Technology devised to root out and understand fake news, how it spreads,

有的科技被开发出来， 用于拔除和理解虚假新闻，

了解它们的传播方式， 并试图降低其扩散。

and to try and dampen its flow.

Humans have to be in the loop of this technology,

人类需要跟进这种科技，

because we can never escape

因为我们无法逃避的是，

that underlying any technological solution or approach

在任何科技解答或手段的背后

is a fundamental ethical and philosophical question

都有一个根本的伦理与哲学问题：

about how do we define truth and falsity,

我们如何定义真实和虚伪，

to whom do we give the power to define truth and falsity

我们将定义真伪的权力托付于谁，

and which opinions are legitimate,

哪些意见是合法的，

which type of speech should be allowed and so on.

哪种言论能被允许， 诸如此类。

Technology is not a solution for that.

科技并非对这个问题的解答,

Ethics and philosophy is a solution for that.

伦理学和哲学才是。

Nearly every theory of human decision making,

人类决策、人类合作和人类协调

human cooperation and human coordination

的几乎每一个理论，

has some sense of the truth at its core.

其核心都存在某种程度的真相。

But with the rise of fake news,

但随着虚假新闻、

the rise of fake video,

虚假视频、

the rise of fake audio,

虚假音频的崛起，

we are teetering on the brink of the end of reality,

我们正在现实终结 的边缘摇摇欲坠，

where we cannot tell what is real from what is fake.

在这里我们无法分辨 何为真实，何为虚假。

And that's potentially incredibly dangerous.

这有可能是极度危险的。

We have to be vigilant in defending the truth

我们必须保持警惕，拒绝虚假信息，

against misinformation.

捍卫真相——

With our technologies, with our policies

通过我们的技术，我们的政策，

and, perhaps most importantly,

以及，或许也是最重要的，

with our own individual responsibilities,

通过我们自己的责任感、

decisions, behaviors and actions.

决定、行为，和举动。

Thank you very much.

谢谢大家。

(Applause)

（掌声）